The record: counted, never forecast

We don’t predict markets. We keep receipts on them.

The market-minute

The unit of the record is one market-minute: one market, one minute of its life. For every market-minute, EdgeDepth stores versioned readings: things like VPIN, book imbalance, open interest, funding and the state of the tape, each measured at that minute. Stack the minutes up and you get something you can ask questions of: every time these readings looked like this, what happened next?

The nightly rebuild and the record edge

Every night the record is rebuilt from raw captured market data, and its edge (the most recent finished minute it covers) moves forward. Research answers always come from the record edge backwards, so a result is never half-built data dressed up as history.

Same search, same bytes

Run a search today and it comes back with a reproducibility key. Preserve the result receipt and its pinned artifact. Running the query against a newer record can produce a different revision; the reproducibility reference explains that boundary. A claim you cannot re-run is a story. Everything here is built to be re-run.

Replay makes an example inspectable

Where the recorded streams and your access support replay, inspect the matching minute: the same trades, the same order book, tick by tick. When a scan says “this happened 89 times”, you can watch any of the 89 yourself rather than take the sentence on faith.

What we refuse to claim

The record supports exactly one kind of sentence: this condition was true N times, and here is what followed each time, out of how many chances. It does not support “this will happen again”, and we do not sell that sentence.

We refuse to claim what the record cannot support. It tells you what happened, out of how many chances. It never tells you what happens next.

See it live

Ask the record a question yourself.

Related: Base rates · Percentiles and the P-chip